Exposure Modelling through Machine Learning for Multi-Hazard Risk Assessment

Abstract

In the last decades, most efforts to catalogue and characterise the built environment for multi-hazard risk assessment have focused on the exploration of housing census data, cadastral datasets, and local surveys. The first approach is only updated every 10 years and does not provide building locations, the second type of data is only available for some urban centres, and the third approach requires surveyors with an engineering background, which is cost-prohibitive for large-scale risk studies. It is thus clear that methods to characterise the built environment for large-scale risk analysis at the asset level are currently missing, which hampers the assessment of the impact of natural hazards for the purposes of risk management. Some recent efforts have demonstrated how machine learning algorithms can be trained to recognise specific architectural and structural features of buildings based on their fa�ades, and probabilistically propose one or multiple building classes. In this study, we demonstrate how such algorithms can be combined with data from OpenStreetMap and imagery from GoogleStreetView to develop exposure models for multi-hazard risk analysis. We developed a training dataset with 5000 images of buildings from the parish of Alvalade, within the Lisbon district (Portugal). This dataset was used to test different algorithms, which led to distinct performance and accuracy levels. The resulting exposure model was used to estimate the impact of an earthquake scenario, and the results were compared with the impact calculated using the ground truth data. We discuss how the uncertainty from such models can be propagated into the risk results, and how this approach can be used for the assessment losses due to hydrological and meteorological hazards.